A build is a moment. An AI system keeps running after you stop looking at it, and that is exactly when it starts to slip. The intake form that sorted leads perfectly in March quietly mislabels them by June, because the questions people ask have changed. Nobody notices until a deal lands in the wrong pile.
This retainer pays for the part nobody puts in the quote. It is not a bug-fix contract. It gives the system an owner after launch.
What you get every month
- It becomes my problem, not yours. One live AI system (an AI agent, an automation, an intake form, or a scheduled job) that I keep running, updated, and on the right AI model.
- A real audit, not a status light. Once a month I check what it actually produced and tell you where it was wrong and why. A dashboard that says "healthy" is the thing that hides the problem.
- One new thing a quarter. One new automation or capability, scoped with you, so the system is more useful in month nine than it was at launch.
- You hear about it first. When a provider changes a model, or your data shifts under the system, that is a message from me, not a surprise from a client.
Where this already runs
On an agency portal, AI sits at three points: an intake form that picks its next question from what the prospect already said, task briefs written from the project scope, and a nightly check of the activity log that emails a person when something looks off. None of them publishes, approves, or sends anything to a client on its own. Keeping that true through four years of changes is the job.
On a healthcare platform, every request to the AI passes through one guard that tells it to refuse protected health information, and free text is checked for it before anything is saved. A rule like that falls apart the month nobody checks, so the version I trust stops a release when someone breaks it. I run 23 checks like that daily against my own platform.
On a second agency system, Claude sorts the in-app feedback queue first, so the founder opens a sorted list instead of a pile. A person still decides.
The gate in front of my own drafting AI is public too. I wrote up why the bottleneck is approval and not generation, including the 41 days this blog went without a post while four drafts sat waiting. That post is the plainest statement of what this retainer sells: the gate, the record, and somebody who reads the output.
More than one system
Two or three systems get one audit across the set, because the problems worth finding sit between them. Shared parts stay on one version, so a fix lands once. The first system to slip warns you about the rest, and the quarterly capability goes to whichever system needs it most. That tier is priced per system, after I see what you are running.
Why one number
One system is $3,500 a month because predictable beats hourly, and the value is in the watching, not the hours. When a build ships, this is the default next step, and you can opt out of it.
I keep these deliberately small, and I run every one of them myself. If your system is stable and genuinely needs nothing, I will tell you to drop the retainer. It has to earn its place every month.
The terms, plainly
- Price. $3,500 a month for one system, starting when the build ships.
- Term. The first 3 months are a commitment. A new system needs that long to settle, and the monthly audit needs a few months of output to compare. After that it runs month to month, and you can stop with 30 days' notice.
- AI and hosting costs. The accounts for the AI model, the database and the hosting are in your name, and those providers bill you directly. I don't resell them or add a margin. My fee pays for the work: the audits, the fixes and the new capability each quarter.